|Publication number||US20050005062 A1|
|Application number||US 10/708,242|
|Publication date||Jan 6, 2005|
|Filing date||Feb 18, 2004|
|Priority date||Jul 2, 2003|
|Also published as||CN1320436C, CN1320437C, CN1553346A, CN1558321A, CN1598755A, CN100334567C, US7281072, US8301809, US20050005044, US20050005063, US20050015545, US20130013827|
|Publication number||10708242, 708242, US 2005/0005062 A1, US 2005/005062 A1, US 20050005062 A1, US 20050005062A1, US 2005005062 A1, US 2005005062A1, US-A1-20050005062, US-A1-2005005062, US2005/0005062A1, US2005/005062A1, US20050005062 A1, US20050005062A1, US2005005062 A1, US2005005062A1|
|Inventors||Ling-Yi Liu, Tse-Han Lee, Michael Gordon Schnapp, Yun-Huei Wang, Chung-Hua Pao|
|Original Assignee||Ling-Yi Liu, Tse-Han Lee, Michael Gordon Schnapp, Yun-Huei Wang, Chung-Hua Pao|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (20), Referenced by (44), Classifications (14), Legal Events (3)|
|External Links: USPTO, USPTO Assignment, Espacenet|
1. Field of the Invention
The invention relates to a redundant external storage virtualization computer system. More particularly, a redundant external storage virtualization computer system that uses point-to-point serial-signal interconnects as the primary device-side IO device interconnects is disclosed.
2. Description of the Prior Art
Storage virtualization is a technology that has been used to virtualize physical storage by combining sections of physical storage devices (PSDs) into logical storage entities, herein referred to as logical media units (LMUs), that are made accessible to a host system. This technology has been used primarily in redundant arrays of independent disks (RAID) storage virtualization, which combines smaller physical storage devices into larger, fault tolerant, higher performance logical media units via RAID technology.
A Storage virtualization Controller, abbreviated SVC, is a device the primary purpose of which is to map combinations of sections of physical storage media to logical media units visible to a host system. IO requests received from the host system are parsed and interpreted and associated operations and data are translated into physical storage device IO requests. This process may be indirect with operations cached, delayed (e.g., write-back), anticipated (read-ahead), grouped, etc. to improve performance and other operational characteristics so that a host IO request may not necessarily result directly in physical storage device IO requests in a one-to-one fashion.
An External (sometimes referred to as “Stand-alone”) Storage Virtualization Controller is a Storage Virtualization Controller that connects to the host system via an IO interface and that is capable of supporting connection to devices that reside external to the host system and, otherwise, operates independently of the host.
One example of an external Storage Virtualization Controller is an external, or stand-alone, direct-access RAID controller. A RAID controller combines sections on one or multiple physical storage devices (PSDs), the combination of which is determined by the nature of a particular RAID level, to form logical media units that are contiguously addressable by a host system to which the logical media unit is made available. A single RAID controller will typically support multiple RAID levels so that different logical media units may consist of sections of PSDs combined in different ways by virtue of the different RAID levels that characterize the different units.
Another example of an external Storage Virtualization Controller is a JBOD emulation controller. A JBOD, short for “Just a Bunch of Drives”, is a set of physical DASDs that connect directly to a host system via one or more a multiple-device IO device interconnect channels. PSDs that implement point-to-point IO device interconnects to connect to the host system (e.g., Parallel ATA HDDs, Serial ATA HDDs, etc.) cannot be directly combined to form a “JBOD” system as defined above for they do not allow the connection of multiple devices directly to the IO device channel.
Another example of an external Storage Virtualization Controller is a controller for an external tape backup subsystem.
The primary motivation in configuring a pair of external storage virtualization controllers (SVCs) into a redundant pair is to allow continued, uninterrupted access to data by a host (or more than one host) even in the event of a malfunction or failure of a single SVC. This is accomplished by incorporating functionality into the SVCs that allow one controller to take over for the other in the event that the other becomes handicapped or completely incapacitated.
On the device side, this requires that both controllers are able to access all of the physical storage devices (PSDs) that are being managed by the SVCs, no matter which SVC any given PSD may initially be assigned to be managed by. On the host side, this requires that each SVC have the ability to present and make available to the host all accessible resources, including those that were originally assigned to be managed by the alternate SVC, in the event that its mate does not initially come on line or goes off line at some point (e.g., due to a malfunction/failure, maintenance operation, etc.).
A typical device-side implementation of this would be one in which device-side IO device interconnects are of the multiple-initiator, multiple-device kind (such as Fibre, Parallel SCSI), and all device-side IO device interconnects are connected to both SVCs such that either SVC can access any PSD connected on a device-side IO device interconnect. When both SVCs are on-line and operational, each PSD would be managed by one or the other SVC, typically determined by user setting or configuration. As an example, all member PSDs of a logical media unit (LMU) that consists of a RAID combination of PSDs would be managed by the particular SVC to which the logical media unit itself is assigned.
A typical host-side implementation would consist of multiple-device IO device interconnects to which the host(s) and both SVCs are connected and, for each interconnect, each SVC would present its own unique set of device IDs, to which LMUs are mapped. If a particular SVC does not come on line or goes off line, the on-line SVC presents both sets of device IDs on the host-side interconnect, its own set together with the set normally assigned to its mate, and maps LMUs to these IDs in the identical way they are mapped when both SVCs are on-line and fully operational. In this kind of implementation, no special functionality on the part of the host that switches over from one device/path to another is required to maintain access to all logical media units in the event that an SVC is not on-line. This kind of implementation is commonly referred to as “transparent” redundancy.
Redundant SVC configurations are typically divided into two categories. The first is “active-standby” in which one SVC is presenting, managing, and processing all IO requests for all logical media units in the storage virtualization subsystem (abbreviated SVS) while the other SVC simply stands by ready to take over in the event that the active SVC becomes handicapped or incapacitated. The second is “active-active” in which both SVCs are presenting, managing, and processing IO requests for the various LMUs that are present in the SVS concurrently. In active-active configurations, both SVCs are always ready to take over for the other in the event that one malfunctions, causing it to become handicapped or incapacitated. Active-active configurations typically provide better levels of performance because the resources of both SVCs (e.g., CPU time, internal bus bandwidth, etc) can be brought to bear in servicing IO requests rather than the resources of only one SVC.
Another essential element of a redundant storage virtualization system is the ability for each SVC to monitor the status of the other. Typically, this would be accomplished by implementing an inter-controller communications channel (abbreviated ICC) between the two SVCs over which they can exchange the operating status. This communications channel may be dedicated, the sole function of which is to exchange parameters and data relating to the operation of the redundant storage virtualization subsystem, or it can be one or more of the IO device interconnects, host-side or device-side, over which operational parameters and data exchange are multiplexed together with host-SVC or device-SVC IO-request-associated data on these interconnects.
Yet another important element of a redundant storage virtualization system is the ability of one SVC to completely incapacitate the other so that it can completely take over for the other SVC without interference. For example, for the surviving SVC to take on the identity of its mate, it may need to take on the device IDs that the SVC going off line originally presented on the host-side IO device interconnect, which, in turn, requires that the SVC going off line relinquish its control over those IDs.
This “incapacitation” is typically accomplished by the assertion of reset signal lines on the controller being taken off line bringing all externally connected signal lines to a pre-defined state that eliminates the possibility of interference with the surviving SVC. Interconnecting reset lines between the SVCs so that one can reset the other in this event is one common way of achieving this. Another way to accomplish this is to build in the ability of an SVC to detect when itself may be malfunctioning and “kill” itself by asserting its own reset signals (e.g., inclusion of a “watchdog” timer that will assert a reset signal should the program running on the SVC fail to poll it within a predefined interval), bringing all externally connected signal lines to a pre-defined state that eliminates the possibility of interference with the surviving SVC.
Traditionally storage virtualization has been done with Parallel SCSI or Fibre IO device interconnects as the primary device-side IO device interconnects connecting physical storage devices to the storage virtualization controller pair. Both Parallel SCSI and Fibre are multiple-device IO device interconnects. A multiple-device IO device interconnect is a form of IO device interconnect that allows multiple IO devices to be connected directly, meaning without the need for any extra off-device intervening active circuitry, to a single host system or multiple host systems (note that typical FC-ALJBODs do have off-device active circuitry, but the purpose of this circuitry is not to enable the interconnection but rather to accommodate the possibility of a failure in the DASD or a swapping out of the DASD that might cause a fatal break in the IO interconnect). Common examples of multiple-device IO device interconnects are Fibre channel FC-AL and Parallel SCSI. Multiple-device IO device interconnects share bandwidth among all hosts and all devices interconnected by the interconnects.
Please refer to
The configuration shown in
While the configuration depicted in
It is therefore a primary objective of the claimed invention to provide a redundant external storage virtualization computer system using point-to-point serial-signal transmissions as the primary device-side IO device interconnects to solve the above-mentioned problem.
According to the claimed invention, a redundant external storage virtualization computer system is introduced. The redundant external storage virtualization computer system includes a host entity for issuing an IO request, a redundant storage virtualization controller pair coupled to the host entity for performing an IO operation in response to the IO request issued by the host entity, and a plurality of physical storage devices for providing storage to the computer system. Each of the physical storage devices is coupled to the redundant storage virtualization controller pair through a point-to-point serial-signal interconnect. The redundant storage virtualization controller pair includes a first and a second storage virtualization controller coupled to the host entity. In the redundant storage virtualization controller pair, when the first storage virtualization controller is not on line or not in operation, the second storage virtualization controller will take over the functionality originally performed by the first storage virtualization controller. In one embodiment, of the present invention, the point-to-point serial-signal interconnect is a Serial ATA IO device interconnect.
It is an advantage of the claimed invention that in the redundant external storage virtualization computer system using Serial ATA as the primary device-side IO device, each physical storage device has a dedicated interconnect to the storage virtualization controller pair.
It is another advantage of the claimed invention that not only the payload data portion of information but also the control information are protected by the SATA IO device interconnect.
These and other objectives of the present invention will no doubt become obvious to those of ordinary skill in the art after reading the following detailed description of the preferred embodiment that is illustrated in the various figures and drawings.
Please refer to
Although there is illustrated in
In order to allow both controllers to access the same PSD 420, an access control switch 342 is inserted in the device-side IO device interconnect path between the SVCs 200 and a PSD 420. Because of the point-to-point nature of the interconnect, only one SVC 200, i.e. the SVC 200 to which the particular PSD 420 is assigned at the time, can be actively accessing the PSD 420 at a time. The other SVC 200 remains in a stand-by mode with respect to this PSD 420 with its IO device interconnect to the particular PSD 420 disabled. One signal line from each SVC 200 is provided for controlling the access control switch 342. The switch 342 determines which SVC interconnect is patched through to the PSD 420.
As depicted in
In a redundant SVS in which it is important that any active components or groups thereof be positioned in hot-swappable units so that the subsystem does not have to be brought down in order to replace such components should a component failure occur. Such hot-swappable units are typically referred to as “Field Replaceable Units” (abbreviated FRU). Being active components, both the PSD itself and the access control switch would quite naturally also be located on FRUs. It makes sense to put them together on the same FRU, for one cannot achieve its intended functionality without the other. Therefore, the access control switch would typically be situated with the PSD in the removable PSD canister.
In one implementation, all of the PSDs 420 in the SVS 20 can be combined to form a PSD array 400, and all the access control switches 342 can be combined to form a switching circuit 340. An example of such implementation is shown in
The host-side IO device interconnect controller 220 is connected to the host entity 10 and the CPC 240 to serve as an interface and buffer between the SVC1 200 and the host entity 10, and receives IO requests and related data from the host entity 10 and maps and/or transfers them to the CPC 240. The host-side IO device interconnect controller 220 comprises one or more host-side ports for coupling to the host entity 10. Some common port types that might be incorporated here are: Fibre Channel supporting Fabric, point-to-point, public loop and/or private loop connectivity in target mode, parallel SCSI operating in target mode, ethernet supporting the iSCSI protocol operating in target mode, Serial-Attached SCSI (SAS) operating in target mode, and Serial ATA operating in target mode.
When the CPC 240 receives the IO requests of the host entity 10 from the host-side IO device interconnect controller 220, CPC 240 parses it and performs some operations in response to the IO requests and sends the data requested and/or reports and/or information of the SVC1 200 back to the host entity 10 through the host-side IO device interconnect controller 220.
After parsing a request received from the host entity 10, while a read request being received and performing one or more operations in response, the CPC 240 get the requested data either internally or from the memory 280, or in both ways, and transfers them to the host entity 10. If the data is not available either internally or does not exists in the memory 280, the IO request will be issued to the PSD array 400 through the SATA IO device interconnect controller 300 and the switching circuit. The requested data will then be transferred from the PSD array 400 to the memory 280 and passed to the host entity 10 through host-side IO device interconnect controller 220.
When a write request is received from the host entity 10, after parsing the request and performing one or more operations, the CPC 240 gets the data from the host entity 10 through the host-side IO device interconnect controller 220, stores them in the memory 280, and then transmits the data to the PSD array 400 through the CPC 240. When the write request is a write back request, the IO complete report can be issued to the host entity 10 first and then the CPC 240 performs the actual write operation later; otherwise, an IO complete report can be issued to the host entity 10 after the requested data is actually written into the PSD array 400.
The memory 280 is connected to the CPC 240 and acts as a buffer to buffer the data transferred between the host entity 10 and the PSD array 400 through the CPC 240. In one embodiment, the memory 280 can be a DRAM; or more particularly, the DRAM can be a SDRAM.
The SATA IO device interconnect controller 300 is the device-side IO device interconnect controller connected between the CPC 240 and the PSD array 400. It serves as an interface and buffer between the SVC 200 and the PSD array 400 and receives IO requests and related data issued from CPC 240 and maps and/or transfers them to the PSD array 400. The SATA IO device interconnect controller 300 re-formats the data and control signals received from CPC 240 to comply with SATA protocol and transmits them to the PSD array 400.
An enclosure management service (EMS) circuitry 360 can be attached to the CPC 240 for management circuitry on an enclosure for containing the PSD array 400. In another arrangement of the SVS 20, the EMS circuitry 360 can be omitted, depending on the actual requirements of the various product functionality. Alternatively, the function of the EMS circuitry 360 can be incorporated into the CPC 240.
In this embodiment, the RCC interconnect controller 236 is implemented in SVC1 200 to connect the CPC 240 to SVC2 200. In addition, the SATA IO device interconnect controller 300 is connected to the PSD array 400 through the switching circuit 340. The switching circuit 340 is also connected to the SVC2 200. In this arrangement, the SVC2 200 can be attached to the SVC1 200. The PSD array 400 can be accessed by the two SVCs 200 through the switching circuit 340. Moreover, the control/data information from the host 10 can be transferred from the CPC 240 through the RCC interconnect controller 236 to the SVC2 200 and further to a second PSD array (not shown).
Data and control signals from host-side IO device interconnect controller 220 enter CPU chip/parity engine 244 through PCI interface 930 and are buffered in PM FIFO 934. The PCI interface 930 to the host-side IO device interconnect controller 220 can be, for example, of a bandwidth of 64-bit, 66 Mhz. When in the PCI slave cycle, the PCI interface 930 owns the PM bus 950 and the data and control signals in the PM FIFO 934 are then transmitted to either the memory interface 920 or to the CPU interface 910.
The data and control signals received by the CPU interface 910 from PM bus 950 are transmitted to CPU 242 for further treatment. The communication between the CPU interface 910 and the CPU 242 can be performed, for example, through a 64 bit data line and a 32 bit address line. The data and control signals can be transmitted to the memory interface 920 through a CM FIFO 922 of a bandwidth of 64 bit, 133 MHz.
An ECC (Error Correction Code) circuit 924 is also provided and connected between the CM FIFO 922 and the memory interface 920 to generate ECC code. The ECC code can be generated, for example, by XORing 8 bits of data for a bit of ECC code. The memory interface 920 then stores the data and ECC code to the memory 280, for example, an SDRAM. The data in the memory 280 is transmitted to PM bus 950 through the ECC correction circuit 926 and compared with the ECC code from the ECC circuit 924. The ECC correction circuit 926 has the functionality of one-bit auto-correcting and multi-bit error detecting.
The parity engine 260 can perform parity functionality of a certain RAID level in response to the instruction of the CPU 242. Of course, the parity engine 260 can be shut off and perform no parity functionality at all in some situation, for example, in a RAID level 0 case. In one embodiment as shown in the
The PLL (Phase Locked Loop) 980 is provided for maintaining desirable phase shifts between related signals. The timer controller 982 is provided as a timing base for various clocks and signals. The internal registers 984 are provided to register status of CPU chip/parity engine 244 and for controlling the traffic on the PM bus 950. In addition, a pair of UART functionality blocks 986 are provided so that CPU chip/parity engine 244 can communicate with outside through RS232 interface.
In an alternative embodiment, PCI-X interfaces can be used in place of the PCI interfaces 930, 932. Those skilled in the art will know such replacement can be easily accomplished without any difficulty.
Please refer to
Next please refer to
The Link layer 700 is then to re-format the FIS into a frame by adding SOF, CRC, EOF, etc., thereto and performing the 8b/10b encoding into encoded 8b/10b characters and transmits it to a PHY layer 710.
The PHY layer 710 will transmit signals through a pair of differential signal lines, transmission lines LTX+, LTX−, to and receive signals through another pair of differential signal lines, reception lines LRX+, LRX−, through the switching circuit 340 to a PSD controller in a PSD 420. The two signal lines of each pair of the signal lines, for example LTX+/LTX−, transmit signals TX+/TX− simultaneously at inverse voltage, for example, +V/−V or V/+V, with respective to a reference voltage Vref so that the voltage difference will be +2v or 2V and thus enhance signal quality. This is also applicable to the transmission of the reception signals RX+/RX− on reception lines LRX+, LRX−.
When receiving a frame from the PHY layer 710, the Link layer 700 will decode the encoded 8b/10b characters and remove the SOF, CRC, EOF. A CRC will be calculated over the FIS to compare with the received CRC to ensure the correctness of the received information. When receiving a FIS from the Link layer 700, the transport layer 690 will determine the FIS type and distribute the FIS content to the locations indicated by the FIS type.
A transmission structure complying with serial ATA protocol is shown in
A primitive consists of a single double-word and is the simplest unit of information that may be communicated between a host and a device. When the bytes in a primitive are encoded, the resulting pattern is not easy to be misinterpreted as another primitive or a random pattern. Primitives are used primarily to convey real-time state information, to control the transfer of information and to coordinate communication between the host and the device. The first byte of a primitive is a special character.
A frame consists of a plurality of double-words, and starts with an SOF (Start Of Frame) primitive and ends with an EOF (End Of Frame) primitive. The SOF is followed by a user payload called a FIS (Frame Information Structure). A CRC (Cyclic-Redundancy Check Code) is the last non-primitive double-word immediately proceeding the EOF primitive. The CRC is calculated over the contents of the FIS. Some other flow control primitives (HOLD or HOLDA) are allowed between the SOF and EOF to adjust data flow for the purpose of speed matching.
The transport layer constructs FISs for transmission and decomposes FISs received from the link layer. The transport layer does not maintain context of ATA commands or previous FIS content. As requested, the transport layer constructs an FIS by gathering FIS content and placing them in proper order. There are various types of FIS, two of which are shown in
As shown in
As shown in
In the embodiment of
Please refer back to
A switchover process is depicted in the flow chart in
At this point, the new access owner is free to issue IO requests to the PSD 420 at will. It may keep ownership until switching back of access ownership is requested by the original access owner, following the same procedure as above. Alternately, at some “convenient” opportunity, as when all the IO requests it has to issue for the time being are executed to completion, the new access owner can automatically start the switch-back process by modifying the state of its access control switch control signal to relinquish accessibility to the PSD 420 and issuing an unsolicited acknowledgement to the original access owner informing it that access ownership has been relinquished and the original access owner can now take back access ownership.
Typically, whether the new access owner keeps ownership until a request for switch-back from the original access owner is posted or it automatically transfers ownership back to the original access owner might be fixed by implementation or might be dynamically decided based on such factors as relative frequency of access by the two SVCs 200 and the relative performance impact of keeping ownership versus automatically restoring ownership to the original access owner.
The mechanism of inter-controller communication to achieve the above switchover process can take any of a number of forms. One possible mechanism of communication, referred to here as the Binary Signal Pair Access Ownership Arbitration mechanism, is a pair of “access request” signal lines per SATA IO device interconnect, with each digital binary signal line having one SVC 200 set/clear (active SVC on this signal line) the signal and the other SVC 200 read the state of the signal (passive SVC). One of the access request signal lines has SVC1 200 active and SVC2 200 passive while the other access request signal line has SVC1 200 passive and SVC2 200 active. On the passive side, the SVC 200 can read both the current state of the alternate SVC's access request signal and whether the signal has changed state since last reading. On reading, the latter will be cleared.
In this mechanism, at the outset, one SVC 200 has ownership of the SATA IO device interconnect and has the signal line associated with the IO device interconnect asserted. When the alternate SVC 200 wishes to assume ownership, it becomes the access requester and asserts the signal line on which it is active. It then monitors the signal line on which it is passive, watching for a change in its state indicating that it was deasserted at some point, which, in turn, indicates that the access owner acknowledged its request. At this point, the requesting SVC 200 can take control of the SATA IO device interconnect by altering the state of the switch 342 so that it is patched through to the PSD 420.
The access owner, on the other hand, continuously monitors for assertion the signal line on which it is passive. Following detection of assertion, at a “convenient” time, it starts waiting for any pending IO requests to complete while queuing any new IO requests. When all pending IO requests are complete, it acknowledges the access control request by deasserting the signal line on which it is active. If it wishes access control to be returned, as when there are new queued IO requests to be issued, it reasserts the signal line. The access requester monitors the signal line on which it is passive for a change in state rather than for a deasserted state because the access owner may have deasserted, then immediately asserted the signal line such that the access requester may not have the chance to detect the deasserted state.
A variation on the above Binary Signal Pair Access Ownership Arbitration mechanism for achieving coordinated access ownership transfer would be to implement a pair of HW circuits, referred to here as “Access Ownership Arbitration” circuits (abbreviated AOA), one for each SVC, that indirectly control the access control switch control signals rather than those signals being controlled directly by the SVC.
The output from one of the two AOAS would be connected to and control the access control switch control signal associated with one of the SVCs and the output from the other circuit would be connected to and control the access control switch control signal associated with other SVC. In addition, each of these AOAS would have the “Access Ownership Request” signals (abbreviated AOR) from both of the SVCs as inputs. When an SVC does not possess nor is requesting access ownership, its AOR is kept in a deasserted state. While in this state, the output signal of the AOA associated with this SVC is inactive. When the SVC wishes to assume ownership, it asserts its AOR. If the other SVC's AOR is not active, then the AOA associated with the requesting SVC would assert its output signal thereby asserting the access control switch control signal associated with the requesting SVC. If the other SVC's AOR is active, then the requesting SVC's AOA's output remains deasserted until the other SVC's AOR is deasserted, at which time the requesting SVC's AOA output becomes active. The requesting SVC's AOA output then remains active until the requesting SVC's AOR is deasserted, independent of the state of the other SVC's AOR. Typically, the two AOAS would be located in close proximity to the access control switch itself, such as in the PSD canister together with the access control switch.
A facility by which a SVC can determine whether or not it was granted access ownership in the event that both SVC's assert their AORs concurrently and by which the SVC currently possessing access ownership can determine when the other SVC is requesting access ownership is also required in this mechanism. Providing a SVC the ability to determine the state of the access control switch would accomplish the former while providing the ability to determine the state of the other SVC's AOR would achieve the latter. However, since these two determinations are made at different times during the access ownership transfer process, they can be combined into a single facility consisting single digital binary signal per SVC, referred to here as the “Alternate SVC Access Ownership Request” signal (abbreviated ASAOR), the state of which can be read by the firmware running on the SVC. Normally, this signal would reflect the state of the other SVC's AOR. However, when the SVC is granted access ownership, its ASAOR would cleared to inactive, independent of the state of the other SVC's AOR, and remain in that state until read by the SVC firmware, after which it would go back to reflecting the state of the other SVC's AOR.
In this mechanism, when a SVC wishes to assume ownership, it asserts it's AOR and then starts monitoring its ASAOR. When the SVC detects that its ASAOR is inactive, it knows it has been granted ownership and can proceed to access the PSD. It then keeps its AOR asserted until it wishes to relinquish access ownership, at which point it deasserts its AOR. During the period in which the SVC wishes to maintain access ownership, it addition to keeping its AOR asserted, it also monitors its ASAOR for assertion. If it detects assertion indicating that the other SVC wishes to assume ownership, at a “convenient” time, it would start waiting for any pending IO requests to complete while queuing any new IO requests. When all pending IO requests are complete, it would then relinquish ownership and deassert its AOR. If it wishes access ownership to be returned, as when there are new queued IO requests to be issued, it would immediately reassert its AOR.
Another possible communication mechanism is passing access ownership transfer requests and acknowledgements over communication channels that support the transfer of multiple bits and/or bytes of information. A set of inexpensive dedicated communication channels, such as 12C channels, can be implemented for the purpose of exchanging these requests and acknowledgements. Alternately, the implementation can take advantage of the existing inter-controller communication channel (ICC) that allows the two SVCs 200 in the redundant pair to communicate with each other to exchange these access ownership transfer requests and acknowledges as part of the normal state-synchronization information that gets exchanged between the two SVCs 200.
A condition under which switchover of access ownership would be mandated is when the access owner SVC 200 malfunctions in such a way that the alternate SVC 200 must take over its functions.
The access control switch 342 will remain in this state until the malfunctioning SVC 200 is replaced or brought back on line and requests ownership to be transferred over to it. The state of the access control switch signal line for each controller at reset, power-up, and during initialization remains such as to disable patching through of itself to the PSD 420 to insure that it does not interfere with potentially on-going PSD 420 accesses by the on-line SVC 200 by inadvertently forcing the access control switch 342 into a state that disrupts such accesses.
An alternate method of handling “occasional” access requirements on the part of the SVCs200 that does not normally have access ownership of the PSD 420 is to have the access owner act as an agent for issuing the IO requests that the SVC 200 requiring access, termed access requester, needs to have executed, an operation termed here as “IO Request Rerouting”. This would typically entail transferring all the necessary IO request information to the access owner for it to construct into an IO request to issue to the PSD 420 for the access requester. In addition to the IO request information, the access requester would transfer any payload data to be written to the PSD to the access owner before or during IO request issuance and execution. Any payload data being read from the PSD would be transferred back to the access requester during or after IO request execution. Completion status of the operation, typically information that indicates whether the operation “succeeded” or “failed” and for what reason, would be passed back to the access requester on completion of the IO request execution.
IO request rerouting function can be performed through a redundant IO device interconnect port pair. Each member port of the IO device interconnect port pair is provided on a separate SVC of the redundant SVC pair but connected to the same PSD through the IO device interconnect. A member port of the redundant IO device interconnect port pair can be a device-side IO device interconnect port, or a hybrid IO device interconnect port which acts as a device-side IO device interconnect port for some IO operations and as a host-side IO device interconnect port for other IO operations. In this regard, the member ports of the redundant IO device interconnect port pair could consist of point-to-point IO device interconnect ports, such as Serial ATA interconnects, or alternately, could consist of multiple-device IO device interconnect ports, such as Fibre Channel or Parallel SCSI interconnects. In addition, such IO request rerouting function can also be applied to the situation that IO requests are performed through the complementary expansion ports of the redundant SVC pair. The complementary expansion ports of the redundant SVC pair will be recited in further detail later with
There are a couple of advantages of IO Request Rerouting over actually transferring access ownership back and forth in order to allow both SVCs 200 to have access to each PSD 420. Firstly, because of the nature of Serial ATA protocol, requiring a rather extended “bring-up” on the interconnect in going from a “down” state to an “up” state, there could be a significant latency between when an access requester receives ownership from the access owner and when it can actually start initiating PSD accesses. Secondly, the process of “downing” the SATA interface then bringing it back up again may result in the need for SATA interface circuitry on either side to enter states of abnormal condition handling. Occasionally, because abnormal condition handling procedures are typically not as thoroughly tested as normal condition processing, bugs may appear which may interfere with the successful rebring-up of the interconnect. To minimize this risk, it is good practice to try to minimize the occurrence of what either side would interpret and deal with as abnormal conditions, which, in this case would include minimizing the instances that access ownership needs be transferred.
One limitation of a “pure” Serial ATA SVC in which all of the device-side IO device interconnects are Serial ATA is that the number of PSDs that can be connected is limited by the number of device-side IO device interconnects that can be packed onto a single SVC. Because the SATA specification only allows for maximum signal line lengths of 1.5 m, the PSDs connected to one SVC must be packed close enough so that no signal line length exceeds 1.5 m. A typical SATA storage virtualization subsystem will only provide for connection of a maximum of 16 SATA PSDs because of these limitations. So a “pure” SATA storage virtualization subsystem is unable to match the expandability of a Fibre FC-AL storage virtualization subsystem, which would typically allow for connection of up to 250 PSDs via connection of external expansion chassis on the same set of device-side IO device interconnects.
In order to overcome this limitation, the current invention optionally includes one or more expansion device-side multiple-device IO device interconnects, herein referred to as device-side expansion ports, such as Parallel SCSI or Fibre FC-AL, on the SVC. These interconnects would typically be wired in such a way as to allow external connection of external expansion chassis. These chassis could be simple “native” JBODs of PSDs directly connected to the interconnect without any intervening conversion circuitry or could be intelligent JBOD emulation subsystems that emulate “native” JBODs using a combination of SATA or PATA PSDs and a single or redundant set of SVCs that provide the conversion from the multiple-device IO device interconnect protocol that provides the connection of the JBOD subsystem to the primary storage virtualization subsystem to the device-side IO device interconnect (SATA or PATA) protocol that provides the connection between the JBOD SVC(s) and the PSDs that they manage.
The current invention introduces three possible options for wiring of the device-side expansion ports.
An enhanced such implementation is depicted in
When it becomes necessary to route all IOs from both SVCs' expansion ports to Storage Unit Port 1 due to a break in the interconnect connecting to Storage Unit Port 2 or perhaps a malfunction in Storage Unit Port 2 itself, M1 and M2 would be set to 1 while M3 would remain clear at 0 and M4 and M5 are “Don't Care”. To route all IOs from both SVCs expansion ports to Storage Unit Port 2, M1 and M3 would be set to 1 while M4 and M5 would remain clear at 0 and M2 is “Don't Care”. If SVC1 goes off line, to route IOs from SVC2 directly to Storage Unit Port 2, M1, M4, and M5 would remain clear at 0 while M2 and M3 are “Don't Care”. If, in addition to SVC1 going off line there is also a break in the interconnect to Storage Unit Port 2 or perhaps a malfunction in the port itself, then IOs from SVC2 would need to be routed to Storage Unit Port 1. This would be done by setting M5 and M2 to 1 while M1 remains clear at 0 and M3 and M4 are “Don't Care”. Conversely, if SVC2 goes off line, to route IOs from SVC1 directly to Storage Unit Port 1, M2 and M3 would remain clear at 0 while M1, M4, and M5 are “Don't Care”. If, in addition to SVC2 going off line there is also a break in the interconnect to Storage Unit Port 1 or perhaps a malfunction in the port itself, then IOs from SVC1 would need to be routed to Storage Unit Port 2. This would be done by setting M3 and M4 to 1 while M5 remains clear at 0 and M1 and M2 are “Don't Care”. The table shown in
The switching could be initiated by a hardware signal detection circuit (SDC) that detects whether or not there is a valid signal present on S1 or S2 or it could be initiated by one of the two SVCs when it detects a break or malfunction in the default path.
Yet another option for wiring of device expansion ports in the configuration depicted in
When an SVC detects that a storage unit connected on an IO device interconnect that connects to one of its expansion ports can no longer be accessed, whether it is due to a detected break/malfunction in the expansion-port/interconnect or some other cause, the detecting SVC passes the IO request to the alternate SVC for the alternate SVC to issue to the same storage unit via the complementary expansion-port/interconnect and alternate storage unit port. Any data/status associated with the IO request is transferred between the two SVCs during the execution of the IO request. If the expansion-port/interconnect on the alternate SVC appears to be up and functioning normally yet access to the storage unit fails on the alternate SVC also, the storage unit would be considered as having failed or having been removed. If access succeeds, then the loss of access would be considered to be localized to the original SVC and IO requests associated with future accesses to the storage unit are automatically rerouted to the alternate SVC for issuance over the complementary expansion-port/interconnect. During this time, the original SVC monitors the accessibility of the storage unit via its expansion-port/interconnect typically by periodically issuing internally generated IO requests that check the state of the interconnect and storage unit. If, at some point, the original SVC discovers that the storage unit can now be accessed over its expansion-port/interconnect, it will stop rerouting IO requests to the alternate SVC and start issuing them directly over its own expansion-port/interconnect again.
Another feature that an SVC might typically implement is redundancy in the host-side interconnects in which multiple host-side interconnect ports are included on the SVC and LMUs are presented to the host identically over two or more of these interconnects. This feature is designed to allow the host the ability to maintain access to the LMU even if one of the interconnects and/or ports on the interconnect should break, become blocked, or otherwise malfunction.
Under normal operation, host(s) can access logical media units through an SVC that is configured to present the LMU over a host-side interconnect. This can be one SVC or both of the SVCs in the redundant pair. If one SVC were to malfunction, logical media units that were already being presented to the host(s) by both SVCs would remain accessible through the normally-functioning SVC and, with the help of special purpose “multiple-redundant-pathing” functionality on the host, on detection that IO request processing through one of the SVCs is disrupted, the IO requests would be completely routed to the normally-functioning SVC.
Those LMUs that were originally only being presented to the host by the SVC that is now malfunctioning would immediately be presented to the host(s) by the normally-functioning SVC over host-side interconnects that connect it to the hosts. For these LMUs, the normally-functioning SVC would be able to transparently take over the processing of host IO requests simply by presenting itself on each interconnect, together with all the reassigned logical media units, in an identical way to what the malfunctioning SVC did prior to its malfunctioning. With this kind of “transparent takeover”, the host need not implement special functionality to make it aware of the SVC malfunctioning and reroute IOs itself in response.
In addition to SVC redundancy, the two sets of complementary ports in turn form a redundant port complement. A host that has two independent ports connected using two separate IO device interconnects to these two complementary redundant port sets then has two independent paths to each logical media unit over which it can issue IO requests. Should one of the ports on the host or on an SVC malfunction or should the IO device interconnect itself break or become blocked, the hosts implementing multiple-redundant-pathing functionality can reroute IO requests over the other redundant path. Alternately, when both paths are functioning normally, the host can elect to issue IO requests over both paths in an effort to balance the load between the paths, a technique referred to as “load balancing”.
To achieve the transparent takeover functionality described above, each of the pair of ports, one on each SVC, that form a complementary port pair are physically interconnected. For bus-style multiple-device IO device interconnects such as Parallel SCSI, the interconnection simply consists of wiring the devices together directly without any intervening circuitry. For other types of interconnects, special switch circuitry may be required to achieve the physical interconnection required.
In configurations in which the hosts implement multiple-redundant-pathing functionality, there is an alternate host-side interconnect configuration that requires fewer interconnects to achieve similar levels of redundancy as shown in
A variation on the Redundant Serial ATA storage virtualization Subsystem uses Parallel ATA PSDs rather than Serial ATA PSDs. For each PSD, it incorporates a SATA-to-PATA conversion circuit that resides in close proximity to the PATA PSD between the access control switch and the PSD and typically together with the access control switch in the same FRU. This conversion circuit converts SATA signals and protocol to PATA and back again in the opposite direction. The importance of a Redundant Serial ATA SVS that uses Parallel ATA PSDs lies in the fact that, in the short term, supplies of Serial ATA drives will still be relatively short compared to Parallel ATA and Serial ATA drives will still be significantly more expensive. During this transitional period, this kind of a subsystem would allow PATA PSDs to be substituted for SATA PSDs, eliminating the concerns over SATA PSD supply and cost. Such a subsystem would typically place the conversion circuit, together with the access control switch, in the removable canister in which the PSD resides. The removable canister allows the PSD and any related circuitry to be easily swapped out in the event of that a PSD and/or related circuitry needs servicing. By placing the conversion circuit in the canister, when SATA drives become readily available at a competitive price point, the entire canister contents could be swapped out with a SATA PSD and SATA related circuitry.
Please refer to
Those skilled in the art will readily observe that numerous modifications and alternations of the device may be made while retaining the teaching of the invention. Accordingly, the above disclosure should be construed as limited only by the metes and bounds of the appended claims.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5928367 *||Apr 29, 1996||Jul 27, 1999||Hewlett-Packard Company||Mirrored memory dual controller disk storage system|
|US5987566 *||Oct 6, 1998||Nov 16, 1999||Emc Corporation||Redundant storage with mirroring by logical volume with diverse reading process|
|US6061752 *||Jun 15, 1998||May 9, 2000||Compaq Computer Corporation||Technique for hot plugging a peripheral controller card containing PCI and SCSI buses on a single connector into a host system board|
|US6483107 *||May 11, 1999||Nov 19, 2002||Josef Rabinovitz||Canister having a combined guide rail and light pipe system for use in a computer peripheral enclosure|
|US6532547 *||Jun 16, 1995||Mar 11, 2003||Emc Corporation||Redundant peripheral device subsystem|
|US6961813 *||Feb 25, 2003||Nov 1, 2005||Hewlett-Packard Development Company, L.P.||System and method for providing multi-initiator capability to an ATA drive|
|US7065661 *||Dec 16, 2002||Jun 20, 2006||Emc Corporation||Using request and grant signals to read revision information from an adapter board that interfaces a disk drive|
|US20020152355 *||Apr 16, 2001||Oct 17, 2002||International Business Machines Corporation||System apparatus and method for storage device controller-based message passing having effective data channel bandwidth and controller cache memory increase|
|US20030033477 *||Jun 28, 2002||Feb 13, 2003||Johnson Stephen B.||Method for raid striped I/O request generation using a shared scatter gather list|
|US20030193776 *||Mar 20, 2003||Oct 16, 2003||Bicknell Bruce A.||Disc storage subsystem having improved reliability|
|US20040177218 *||Nov 5, 2003||Sep 9, 2004||Meehan Thomas F.||Multiple level raid architecture|
|US20040210584 *||Feb 25, 2004||Oct 21, 2004||Peleg Nir||Method and apparatus for increasing file server performance by offloading data path processing|
|US20050005044 *||Jan 20, 2004||Jan 6, 2005||Ling-Yi Liu||Storage virtualization computer system and external controller therefor|
|US20050089027 *||Jun 18, 2003||Apr 28, 2005||Colton John R.||Intelligent optical data switching system|
|US20050120170 *||Dec 2, 2003||Jun 2, 2005||Nvidia Corporation||Universal raid class driver|
|US20050160319 *||Jan 14, 2004||Jul 21, 2005||Marcak Scott B.||System and method for detecting and reporting resource conflicts|
|US20050182887 *||Feb 17, 2004||Aug 18, 2005||Toshimi Sakurai||PCI-express to PCI/PCI X translator|
|US20050204078 *||Dec 13, 2004||Sep 15, 2005||Steinmetz Joseph H.||Integrated-circuit implementation of a storage-shelf router and a path controller card for combined use in high-availability mass-storage-device shelves that may be incorporated within disk arrays|
|US20050223269 *||Mar 14, 2005||Oct 6, 2005||Stolowitz Michael C||Disk controller methods and apparatus with improved striping, redundancy operations and interfaces|
|US20050251588 *||Mar 26, 2002||Nov 10, 2005||Genx Systems, Inc.||Method and apparatus for supporting access of a serial ATA storage device by multiple hosts with separate host adapters|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7124143||Jun 28, 2004||Oct 17, 2006||Hitachi, Ltd.||Data migration in storage system|
|US7127558||Jan 2, 2004||Oct 24, 2006||Hitachi, Ltd.||Virtualization controller, access path control method and computer system|
|US7334029||Nov 4, 2004||Feb 19, 2008||Hitachi, Ltd.||Data migration method|
|US7363457||Jul 21, 2005||Apr 22, 2008||Sun Microsystems, Inc.||Method and system for providing virtualization data services for legacy storage devices|
|US7536584||Oct 23, 2006||May 19, 2009||Dot Hill Systems Corporation||Fault-isolating SAS expander|
|US7673185||Aug 29, 2007||Mar 2, 2010||Dot Hill Systems Corporation||Adaptive SAS PHY configuration|
|US7912814||Sep 23, 2005||Mar 22, 2011||Hitachi, Ltd.||Data migration in storage system|
|US7934024 *||Nov 28, 2005||Apr 26, 2011||Via Technologies Inc.||Method for reading and writing non-standard register of serial advanced technology attachment (SATA) device|
|US8041862 *||Aug 24, 2010||Oct 18, 2011||Quantum Corporation||Data flow control and bridging architecture enhancing performance of removable data storage systems|
|US8099532||Jun 14, 2007||Jan 17, 2012||International Business Machines Corporation||Intelligent dynamic multi-zone single expander connecting dual ported drives|
|US8141092||Nov 15, 2007||Mar 20, 2012||International Business Machines Corporation||Management of an IOV adapter through a virtual intermediary in a hypervisor with functional management in an IOV management partition|
|US8141093||Nov 15, 2007||Mar 20, 2012||International Business Machines Corporation||Management of an IOV adapter through a virtual intermediary in an IOV management partition|
|US8141094||Dec 3, 2007||Mar 20, 2012||International Business Machines Corporation||Distribution of resources for I/O virtualized (IOV) adapters and management of the adapters through an IOV management partition via user selection of compatible virtual functions|
|US8144582||Dec 30, 2008||Mar 27, 2012||International Business Machines Corporation||Differentiating blade destination and traffic types in a multi-root PCIe environment|
|US8201020||Nov 12, 2009||Jun 12, 2012||International Business Machines Corporation||Method apparatus and system for a redundant and fault tolerant solid state disk|
|US8214569||Feb 10, 2011||Jul 3, 2012||Via Technologies Inc.||Method for reading and writing non-standard register of serial advanced technology attachment (SATA) device|
|US8281044||Aug 22, 2011||Oct 2, 2012||Quantum Corporation||Data flow control and bridging architecture enhancing performance of removable data storage systems|
|US8285893 *||Oct 13, 2006||Oct 9, 2012||Dell Products L.P.||System and method for adaptively setting connections to input/output hubs within an information handling system|
|US8296476||Jun 24, 2011||Oct 23, 2012||Chengdu Huawei Symantec Technologies Co., Ltd.||Storage method, storage system, and controller|
|US8301810||Oct 11, 2005||Oct 30, 2012||Infortrend Technology, Inc.||SAS storage virtualization controller, subsystem and system using the same, and method therefor|
|US8352649 *||Jun 7, 2006||Jan 8, 2013||Infortrend Technology, Inc.||Storage virtualization subsystem architecture|
|US8352653||Nov 22, 2006||Jan 8, 2013||Infortrend Technology, Inc.||Storage virtualization subsystem and system with host-side redundancy via SAS connectivity|
|US8359415 *||May 5, 2008||Jan 22, 2013||International Business Machines Corporation||Multi-root I/O virtualization using separate management facilities of multiple logical partitions|
|US8427941||Sep 22, 2006||Apr 23, 2013||Infortrend Technology, Inc.||Redundant storage virtualization subsystem having data path branching functionality|
|US8489914||Apr 30, 2012||Jul 16, 2013||International Business Machines Corporation||Method apparatus and system for a redundant and fault tolerant solid state disk|
|US8533409 *||Jan 25, 2007||Sep 10, 2013||Infortrend Technology, Inc.||Method of managing data snapshot images in a storage system|
|US8756454||May 13, 2013||Jun 17, 2014||International Business Machines Corporation||Method, apparatus, and system for a redundant and fault tolerant solid state disk|
|US20050050271 *||Jan 2, 2004||Mar 3, 2005||Kiyoshi Honda||Virtualization controller, access path control method and computer system|
|US20060136666 *||Oct 11, 2005||Jun 22, 2006||Ching-Te Pang||SAS storage virtualization controller, subsystem and system using the same, and method therefor|
|US20060136688 *||Oct 11, 2005||Jun 22, 2006||Ching-Te Pang||Redundant SAS storage virtualization subsystem and system using the same, and method therefor|
|US20060282639 *||Jun 7, 2006||Dec 14, 2006||Infortrend Technology Inc.||Storage virtualization subsystem architecture|
|US20060293764 *||Nov 28, 2005||Dec 28, 2006||Jar-Haur Wang||Method for reading and writing non-standard register of serial advanced technology attachment (SATA) device|
|US20070070887 *||Sep 22, 2006||Mar 29, 2007||Infortrend Technology Inc.||Redundant storage virtualization subsystem having data path branching functionality|
|US20070174569 *||Jan 25, 2007||Jul 26, 2007||Infortrend Technology, Inc.||Method of managing data snapshot images in a storage system|
|US20140032960 *||May 9, 2013||Jan 30, 2014||Fujitsu Limited||Information processing system and access control method|
|CN102255958A *||Jun 27, 2011||Nov 23, 2011||北京星网锐捷网络技术有限公司||Data synchronization method and system thereof|
|EP1768026A1 *||Sep 22, 2006||Mar 28, 2007||Infortrend Technology, Inc.||Redundant storage virtualization subsystem having data path branching functionality|
|EP1796003A2 *||Nov 16, 2006||Jun 13, 2007||Infortrend Technology, Inc.||Storage virtualization subsystem and system with host-side redundancy via sas connectivity|
|EP1816563A2 *||Jan 25, 2007||Aug 8, 2007||Infortrend Technology, Inc.||Method of managing data snapshot images in a storage system|
|EP1816563A3 *||Jan 25, 2007||Sep 22, 2010||Infortrend Technology, Inc.||Method of managing data snapshot images in a storage system|
|EP2000911A1 *||Mar 31, 2008||Dec 10, 2008||Reven.GE@S.R.L.||Computer system comprising at least two computers, and related method for continuous operation of said system|
|EP2378430A1 *||Nov 27, 2009||Oct 19, 2011||Chengdu Huawei Symantec Technologies Co., Ltd.||Storing method, storing system and controller|
|WO2007146515A2 *||May 9, 2007||Dec 21, 2007||Dot Hill Systems Corp||Fault-isolating sas expander|
|WO2011057885A1 *||Oct 21, 2010||May 19, 2011||International Business Machines Corporation||Method and apparatus for failover of redundant disk controllers|
|International Classification||G06F13/00, G06F13/14, G06F12/08, G06F3/06, G06F12/00|
|Cooperative Classification||G06F3/0683, G06F3/0658, G06F3/0664, G06F3/0607|
|European Classification||G06F3/06A4V2, G06F3/06A2A4, G06F3/06A4T4, G06F3/06A6L4|
|Feb 18, 2004||AS||Assignment|
Owner name: INFORTREND TECHNOLOGY, INC., TAIWAN
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:LIU, LING-YI;LEE, TSE-HAN;SCHNAPP, MICHAEL GORDON;AND OTHERS;REEL/FRAME:014345/0395;SIGNING DATES FROM 20031218 TO 20040116
|Mar 29, 2011||FPAY||Fee payment|
Year of fee payment: 4
|Feb 10, 2015||FPAY||Fee payment|
Year of fee payment: 8